I think this would be a handy feature to have and would be very useful.
I've been thinking quite a bit about expanding the definition of content. I'm putting together some slides this week (hopefully) that will explain this thinking a bit more..
In short the concept is that content means more than just text and front matter meta data. Images are content and have their own meta data. It would be great if they were treated as such. All of the rules around nodes & pages would be extended to these alternative content types.
Ultimately the difference in these two approaches is that the readDir / readImage approach works outside of what Hugo already gives you where the approach I'm thinking works inside of Hugo..
I admit it's a bit rough today and there's no reason to suppose that both approaches shouldn't be used. Just as ReadDir works alongside the Hugo content processing today.
Perhaps just seeing what I'm thinking would help you design your feature in a way that could be leveraged further in the future. In fact, the more I think about it the more I think what you've outlined may be step 1 towards what I'm thinking.